Exposure Correlation & Double Counting
When explaining why univariate relativities (e.g., Pure Premium Method) suffer from double counting when exposures are correlated:
- The Issue: Pure premium calculations assume independent distributions. When exposures are correlated, disproportionate concentrations of high-cost or low-cost classes distort the relativities.
- Suggested Phrasing:
“Disproportionate exposures are concentrated in high- and low-cost groups. For example, 80% of Class 99 exposures are concentrated in the high-cost Territory Z, whereas the univariate method assumes a balanced distribution. This causes Class 99 to appear costlier than it actually is, resulting in double counting of the territory effect.”
GLM Modeling
- Signal vs. Noise: Avoid saying GLMs “model noise.”
- Suggested Phrasing:
“GLMs may overfit and reflect the noise component in the data in addition to the true underlying signal.”
Claim Process Changes
When describing the impact of a slowdown or speedup in claims processing on development:
- The Issue: Applying historical development factors to current, changed data levels causes distortion.
- Suggested Phrasing:
“Applying historical development factors to current (lower/higher) levels of paid or reported losses will lead to overestimation or underestimation of ultimate claims.”
Predictiveness of Case Reserves
- Refining Statement: Simplify claims development descriptions.
- Revision:
- Draft: “Case outstanding observed are predictive of unobserved case reserves at later maturities.”
- Better: “Case outstanding observed are predictive of future claim development.”
General Terminology Precision
Be specific in exam responses instead of using generic terms like “results”:
- Instead of “results”, specify exactly what you mean:
- Projections (or ultimate estimates)
- Age-to-age factors (or LDFs)
- Premium trends or loss ratios
- Only use “results” when referring to the overall analysis outcomes generally.